AI Coding Assistants Become an Attack Surface: Symlink Trick Yields RCE Across Six Tools

Adversa AI·medium signal

AI-assisted vulnerability discovery and agent-targeting attacks are an accelerating theme: Adversa AI demonstrated that a symlink-disguised file copy can trick AI coding assistants into remote code execution, testing six major tools that were all vulnerable, while Bruce Schneier published an analysis on how AI is reshaping vulnerability disclosure. For developers running coding agents with file-system access, the takeaway is that the agent itself is now part of the attack surface and needs sandboxing.